Petri net reduction applied to performance modeling

Designers use Petri Nets and queuing systems to model computer systems to explore different design alternatives and to eliminate designs that do not satisfy the performance requirements for the system. Unfortunately, both Petri Nets and queuing systems demand undue expertise from the designer, and neither Petri Nets nor queuing systems is amenable to further refinement towards a practical implementation. Thus, there is a need for a design environment that not only lets a designer model a computer system without being an expert in any complex mathematical theory but also allows refining the performance model to an acceptable implementation that satisfies the original performance, dependability, and cost requirements. Such an environment is the goal of the Center for Semicustom Integrated Systems (CSIS) effort and this thesis.; The CSIS methodology shields the user from mathematical complexity by providing a set of simple modeling constructs called primitive elements using which complex models are built. This thesis formalizes the CSIS methodology by providing a Petri Net (PN) description for each one of the primitive elements. As a result, any model that uses only the predefined elements also defines a PN model for the system under design. Since a PN model tends to be large in terms of the number of nodes, the PN model is hard to simulate and analyze. This thesis proposes a set of reduction rules to reduce the PN model. The reduced PN model not only aids analysis by simplifying the PN model, it also speeds up the simulation of the original system level model. The PN model is simulated by translating the model into an equivalent VHDL model and simulating the VHDL model using any standard VHDL-1076 simulator. The above theoretical advances have been incorporated into the next generation of CSIS methodology and tool. Using this tool, the author was able to obtain significant reductions in terms of the number of PN nodes and speed up in the simulation of the original system level model.